And, on the other hand, if you actually have a specific vision of what you want to create, it is often quite difficult, or actually impossible, to get an AI to generate an output that fulfills your particular vision.
I'm also learning to draw. As a counterpoint, I found it quite fun and enlightening to use artwork I like as practicing material. You can learn a lot by trying to copy parts of an artwork (without tracing and without uploading the result anywhere) : By copying, you still actually have to draw the artwork and deal with the same challenges that the original artist did - however you can use the original artwork as a sort of guide or cheat sheet on how those problems can be addressed.
I found it often even gave me a deeper appreciation for the original artist, because you see all the small, individual decisions they had to make to eventually get to the artwork - it felt sort of similar to reading code and trying to get into the head of the original developers.
You can also get creative and only copy parts of the artwork or stitch together a new piece from different sources - essentially doing by hand what image generators do automatically. (Just don't upload your new artwork anywhere without contacting the original authors first)
I've been an "amateurish" guitar player for 30 years. I play chords, I do some OK picking and I love trying to "play" songs I like, even if badly played (I like speed metal and shredding, which I know I will never master in my lifetime).
Nevertheless, I do it because I like it, I've seen Satriani, Oscar Lopez, Steve Vai play, and I know I'll never be like them. However I don't think "why bother". I enjoy playing basic guitar some afternoons in my living room, and my wife enjoys listening to me while reading a book. I've always played for me first, and because I enjoyed it.
It is a very rare opportunity that we people have the ability/resources to do something just because we enjoy it. With the advent of this wave of AI, we will need to learn to do more things that way.
Same here, it's extremely hard to find motivation when I know every idea I have has been done better by someone who isn't me, or recently by a computer. In either case I can't meaningfully compete and if it's been done better, why do it at all?
This applies to art, but even more broadly than that. I get no satisfaction in creating something when I can see people who have done it far better than me via a quick search.
I think the current popular model works out to somewhere around a byte in the model per sample image.
Here's a diagram https://www.reddit.com/user/AnOnlineHandle made: https://i.redd.it/yxflzg0z6p3a1.png
Hell, I think I as a developer could do that... Gonna invent a new open source license.
If you think you'll get paid, you probably won't because AI trains on very large datasets, and so pretty much any price per image would quickly get too expensive.
If you think it'll hamstring AI, I doubt it'll be for long. People will just end up curating their dataset better, and images that allow for it will continue to be created.
Also, per https://opensource.org/osd -- "No Discrimination Against Fields of Endeavor", so this likely won't get much uptake in open source circles either.
Not being in the dataset maybe? There's many artists that hate that you can just write "in the style of" with their name.
So at least, if it does appear that it works with their name, they now either have been paid that amount, or they can sue for that amount of lost revenue. At least not everything would be lost.
I'm not a lawyer though, so really no idea whether this could even be possible as it's seems like currently it's fair use.
Very few artists have an unique style. Some are very recognizable, yes, but plenty others do more or less the same sort of thing.
There's thousands of photographers out there that go for the "razor sharp, high contrast closeup that shows every pore" style (https://photoaspects.com/wp-content/uploads/2017/09/black-an...) for instance.
With very limited exceptions you could go to a photographer and say "I want a photo that looks like this Vogue cover" and get it done.
> So at least, if it does appear that it works with their name, they now either have been paid that amount, or they can sue for that amount of lost revenue. At least not everything would be lost.
Don't be silly. If that logic worked, Microsoft could have sued Linus Torvalds for providing an OS for free. Or if you want to be a bit more exact, Sun Microsystems, IBM, or any other Unix vendor.
Not to mention, the kid may enjoy making art, and his/her art may give others joy, even if it's not a profitable career. It can still be a fun hobby - synthesizers have been around for nearly half a century now, but people still play guitar :)
> Dancing is one that will be very hard to copy
I'm not sure I'd hang my hat on a particular current inability of AI, particularly inabilities where there's no obvious impediment to AI doing it. You're just begging for AI to eventually gain that ability, and now you're back where you started. Hooking AI up to physical actuators and getting it to work properly strikes me as a "this will 100% happen, I guarantee this is already being worked on" situation; basing your self-worth off the inability of AI to do those things seems like a ticking time bomb.
They don't require mining rare earth metals, they're more versatile, and uh... oh gosh it's only a matter of time before we're all replaced by cheap carbon based robots. It's been real <3
Similarly, people still play chess, go, and Starcraft even though AI can beat them. Human Jeopardy matches still get televised even though Watson could beat them. We still have math olympics even though Wolfram Alpha could have won all of them decades ago.
I think human dancers are safe for as long as there are humans.
For now. There's no technical reason why we can't have CNC painting machines - there's plenty of existing tech in manufacturing that has the required precision and range of motion. I think we don't see such machines because there is no business need for them - or at least, there wasn't.
To date, the kind of digital images you'd want to reproduce using a CNC painter would be drawn by actual artists, using their hands and - most likely - tools like pens and drawing tablets. The work of an artist is similar enough to actual painting that, if there's a need for the 3D texture, they can just... use real brush on real canvas. There technically is a market for cheap reproductions, but AFAIK this one is covered by "painter farms" - warehouses full of low-paid painters mass-producing paintings, in countries with cheap labor costs.
Now, if you could get the ML model produce "original" work, encoded as a time series of brush strokes (either straight from the model, or by another model turning JPEGs into painting instructions), then this just might provide an incentive for CNC painters. It might even start as a hobby thing - there's lots of people who would build such machine, the way they build 3D printers now, so they could produce their own paintings at home. Those same people will be unlikely to pay a far-away painter farm instead.
> Dancing is one that will be very hard to copy, because you need actual humans for it.
AFAIR that was already done, to a passable level, by Boston Dynamics, a few years ago. Sure, that was a one-off carefully choreographed stunt, but it demonstrated that 1) robotics can handle dancing just fine, and 2) the result is actually enjoyable to watch. The missing bit is software, but given the ML work being done in recent years, in particular all those video editing/generating models that start by parsing human movement from input videos, I think it won't be hard for someone to start generating dance moves.
Overall, the easy part of AI art is generic portraits where you either don't care about the details, or want to use existing, popular things. Eg, you can ask for Harry Potter. And you can ask for a random character. But if you have a personal, distinctive design you made and want to bring to life, then it can get very tough pretty fast.
In fact, the best possible user for an AI generator would be an artist that's also capable of training a custom model. That opens up way more possibilities. Here's what this looks like:
https://www.reddit.com/r/NovelAi/comments/z7qat8/ai_as_a_too...